Re: [PATCH] media: rcar-csi2: Fix PHTW table values for E3/V3M

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Niklas,

On Tuesday, 11 December 2018 04:01:15 EET Niklas Söderlund wrote:
> On 2018-12-10 22:16:52 +0200, Laurent Pinchart wrote:
> > On Monday, 10 December 2018 16:53:55 EET Jacopo Mondi wrote:
> >> The PHTW selection algorithm implemented in rcsi2_phtw_write_mbps()
> >> checks for lower bound of the interval used to match the desired
> >> bandwidth. Use that in place of the currently used upport bound.
> > 
> > The rcsi2_phtw_write_mbps() function performs the following (error
> > handling removed):
> > 
> >         const struct rcsi2_mbps_reg *value;
> >         
> >         for (value = values; value->mbps; value++)
> >                 if (value->mbps >= mbps)
> >                         break;
> >         
> >         return rcsi2_phtw_write(priv, value->reg, code);
> > 
> > With this patch, an mbps value of 85 will match the second entry in the
> > phtw_mbps_v3m_e3 table:
> > 
> > [0]	{ .mbps =   80, .reg = 0x00 },
> > [1]	{ .mbps =   90, .reg = 0x20 },
> > ...
> > 
> > The datasheet however documents the range 80-89 to map to 0x00.
> > 
> > What am I missing ?
> 
> I'm afraid you are missing a issue with the original implementation of
> the rcar-csi2 driver (my fault). The issue you point out is a problem
> with the current freq selection logic not the tables themself which
> needs to be corrected.
> 
> This patch aligns the table with the other tables in the driver and is
> sound. A patch (Jacopo care to submit it?) is needed to resolve the
> faulty logic in the driver. It should select the range according to
> Laurents findings and not the range above it as the current code does.

I wonder whether we should instead modify the tables, to avoid making the 
selection logic more complicated and less efficient CPU-wise.

Speaking of which, the tables are interestingly specified in three different 
ways in the datasheet:

- The TESTDIN_DATA table specifies non-overlapping ranges

e.g. [80, 89]: 0x00, [90, 99]: 0x20, [100, 109]: 0x40, ...

In this regard this patch is an improvement (that is if the faulty selection 
logic gets fixed), as otherwise a frequency of 109.5 would be classified in 
the [110, 129] range, while I think it is meant to be in the [100, 109] range.

Another option would be to set the table mbps value to the high bound of the 
range:

        { .mbps =   90, .reg = 0x00 },
        { .mbps =  100, .reg = 0x20 },
        { .mbps =  110, .reg = 0x40 },
        ...

and use strict lower comparison logic:

        const struct rcsi2_mbps_reg *value;

        for (value = values; value->mbps; value++)
                if (mbps < value->mbps)
                        break

- The PHTW table specifies individual bit rates

e.g. 80: 0x86, 90: 0x86, 100: 0x87, ..

I'm not sure how to interpret this. If I had to guess, I would say it means

[80, 90[: 0x86, [90, 100[: 0x86, [100, 110[: 0x87, ..

We could thus use the same logic than for the TESTDIN_DATA table (and while at 
it merge adjacent ranges that share the same PHTW value).

- The HSFREQRANGE tables specify overlapping ranges

e.g. [80, 131.25]: 0x20, [80.75, 141.75]: 0x30, [90.25, 152.25]: 0x01, ...

This has to be converted to non-overlapping ranges. I would advice centering 
each non-overlapping range to the center of the corresponding overlapping 
range. Today we instead have non-overlapping ranges in the driver whose bounds 
are set to the center of the overlapping ranges, and I don't think this is 
right.

And we could then use the same logic as above here too.

The downside is that the tables would need to be carefully reviewed as they 
would derive from the values in the datasheet instead of computing them 
blindly, and the upside would be simpler code. If we want to instead copy the 
tables blindly for ease of review, then I think we'll need more complex 
selection logic, with different logics for the different tables.

> >> Fixes: 10c08812fe60 ("media: rcar: rcar-csi2: Update V3M/E3 PHTW
> >> tables")
> >> Signed-off-by: Jacopo Mondi <jacopo+renesas@jmondi.org>
> >> ---
> >> 
> >>  drivers/media/platform/rcar-vin/rcar-csi2.c | 62 +++++++++++------------
> >>  1 file changed, 31 insertions(+), 31 deletions(-)
> >> 
> >> diff --git a/drivers/media/platform/rcar-vin/rcar-csi2.c
> >> b/drivers/media/platform/rcar-vin/rcar-csi2.c index
> >> 80ad906d1136..7e9cb8bcfe70 100644
> >> --- a/drivers/media/platform/rcar-vin/rcar-csi2.c
> >> +++ b/drivers/media/platform/rcar-vin/rcar-csi2.c
> >> @@ -152,37 +152,37 @@ static const struct rcsi2_mbps_reg
> >> phtw_mbps_h3_v3h_m3n[] = { };
> >> 
> >>  static const struct rcsi2_mbps_reg phtw_mbps_v3m_e3[] = {
> >> -	{ .mbps =   89, .reg = 0x00 },
> >> -	{ .mbps =   99, .reg = 0x20 },
> >> -	{ .mbps =  109, .reg = 0x40 },
> >> -	{ .mbps =  129, .reg = 0x02 },
> >> -	{ .mbps =  139, .reg = 0x22 },
> >> -	{ .mbps =  149, .reg = 0x42 },
> >> -	{ .mbps =  169, .reg = 0x04 },
> >> -	{ .mbps =  179, .reg = 0x24 },
> >> -	{ .mbps =  199, .reg = 0x44 },
> >> -	{ .mbps =  219, .reg = 0x06 },
> >> -	{ .mbps =  239, .reg = 0x26 },
> >> -	{ .mbps =  249, .reg = 0x46 },
> >> -	{ .mbps =  269, .reg = 0x08 },
> >> -	{ .mbps =  299, .reg = 0x28 },
> >> -	{ .mbps =  329, .reg = 0x0a },
> >> -	{ .mbps =  359, .reg = 0x2a },
> >> -	{ .mbps =  399, .reg = 0x4a },
> >> -	{ .mbps =  449, .reg = 0x0c },
> >> -	{ .mbps =  499, .reg = 0x2c },
> >> -	{ .mbps =  549, .reg = 0x0e },
> >> -	{ .mbps =  599, .reg = 0x2e },
> >> -	{ .mbps =  649, .reg = 0x10 },
> >> -	{ .mbps =  699, .reg = 0x30 },
> >> -	{ .mbps =  749, .reg = 0x12 },
> >> -	{ .mbps =  799, .reg = 0x32 },
> >> -	{ .mbps =  849, .reg = 0x52 },
> >> -	{ .mbps =  899, .reg = 0x72 },
> >> -	{ .mbps =  949, .reg = 0x14 },
> >> -	{ .mbps =  999, .reg = 0x34 },
> >> -	{ .mbps = 1049, .reg = 0x54 },
> >> -	{ .mbps = 1099, .reg = 0x74 },
> >> +	{ .mbps =   80, .reg = 0x00 },
> >> +	{ .mbps =   90, .reg = 0x20 },
> >> +	{ .mbps =  100, .reg = 0x40 },
> >> +	{ .mbps =  110, .reg = 0x02 },
> >> +	{ .mbps =  130, .reg = 0x22 },
> >> +	{ .mbps =  140, .reg = 0x42 },
> >> +	{ .mbps =  150, .reg = 0x04 },
> >> +	{ .mbps =  170, .reg = 0x24 },
> >> +	{ .mbps =  180, .reg = 0x44 },
> >> +	{ .mbps =  200, .reg = 0x06 },
> >> +	{ .mbps =  220, .reg = 0x26 },
> >> +	{ .mbps =  240, .reg = 0x46 },
> >> +	{ .mbps =  250, .reg = 0x08 },
> >> +	{ .mbps =  270, .reg = 0x28 },
> >> +	{ .mbps =  300, .reg = 0x0a },
> >> +	{ .mbps =  330, .reg = 0x2a },
> >> +	{ .mbps =  360, .reg = 0x4a },
> >> +	{ .mbps =  400, .reg = 0x0c },
> >> +	{ .mbps =  450, .reg = 0x2c },
> >> +	{ .mbps =  500, .reg = 0x0e },
> >> +	{ .mbps =  550, .reg = 0x2e },
> >> +	{ .mbps =  600, .reg = 0x10 },
> >> +	{ .mbps =  650, .reg = 0x30 },
> >> +	{ .mbps =  700, .reg = 0x12 },
> >> +	{ .mbps =  750, .reg = 0x32 },
> >> +	{ .mbps =  800, .reg = 0x52 },
> >> +	{ .mbps =  850, .reg = 0x72 },
> >> +	{ .mbps =  900, .reg = 0x14 },
> >> +	{ .mbps =  950, .reg = 0x34 },
> >> +	{ .mbps = 1000, .reg = 0x54 },
> >> +	{ .mbps = 1050, .reg = 0x74 },
> >>  	{ .mbps = 1125, .reg = 0x16 },
> >>  	{ /* sentinel */ },
> >>  
> >>  };

-- 
Regards,

Laurent Pinchart